4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * libcfs/include/libcfs/libcfs_private.h
38 * Various defines for libcfs.
42 #ifndef __LIBCFS_PRIVATE_H__
43 #define __LIBCFS_PRIVATE_H__
45 #ifndef DEBUG_SUBSYSTEM
46 # define DEBUG_SUBSYSTEM S_UNDEFINED
54 * When this is on, LASSERT macro includes check for assignment used instead
55 * of equality check, but doesn't have unlikely(). Turn this on from time to
56 * time to make test-builds. This shouldn't be on for production release.
58 #define LASSERT_CHECKED (0)
64 * Strange construction with empty "then" clause is used to trigger compiler
65 * warnings on the assertions of the form LASSERT(a = b);
67 * "warning: suggest parentheses around assignment used as truth value"
69 * requires -Wall. Unfortunately this rules out use of likely/unlikely.
71 #define LASSERTF(cond, fmt, ...) \
76 LIBCFS_DEBUG_MSG_DATA_DECL(__msg_data, D_EMERG, NULL); \
77 libcfs_debug_msg(&__msg_data, \
78 "ASSERTION( %s ) failed: " fmt, #cond, \
80 lbug_with_loc(&__msg_data); \
84 #define LASSERT(cond) LASSERTF(cond, "\n")
86 #else /* !LASSERT_CHECKED */
88 #define LASSERTF(cond, fmt, ...) \
90 if (unlikely(!(cond))) { \
91 LIBCFS_DEBUG_MSG_DATA_DECL(__msg_data, D_EMERG, NULL); \
92 libcfs_debug_msg(&__msg_data, \
93 "ASSERTION( %s ) failed: " fmt, #cond, \
95 lbug_with_loc(&__msg_data); \
99 #define LASSERT(cond) LASSERTF(cond, "\n")
100 #endif /* !LASSERT_CHECKED */
101 #else /* !LIBCFS_DEBUG */
102 /* sizeof is to use expression without evaluating it. */
103 # define LASSERT(e) ((void)sizeof!!(e))
104 # define LASSERTF(cond, ...) ((void)sizeof!!(cond))
105 #endif /* !LIBCFS_DEBUG */
107 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
109 * This is for more expensive checks that one doesn't want to be enabled all
110 * the time. LINVRNT() has to be explicitly enabled by --enable-invariants
113 # define LINVRNT(exp) LASSERT(exp)
115 # define LINVRNT(exp) ((void)sizeof!!(exp))
118 #define KLASSERT(e) LASSERT(e)
120 void lbug_with_loc(struct libcfs_debug_msg_data *) __attribute__((noreturn));
124 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_EMERG, NULL); \
125 lbug_with_loc(&msgdata); \
128 extern atomic_t libcfs_kmemory;
134 # define libcfs_kmem_inc(ptr, size) \
136 atomic_add(size, &libcfs_kmemory); \
139 # define libcfs_kmem_dec(ptr, size) \
141 atomic_sub(size, &libcfs_kmemory); \
144 # define libcfs_kmem_read() \
145 atomic_read(&libcfs_kmemory)
148 # define libcfs_kmem_inc(ptr, size) do {} while (0)
149 # define libcfs_kmem_dec(ptr, size) do {} while (0)
150 # define libcfs_kmem_read() (0)
151 #endif /* LIBCFS_DEBUG */
153 #ifndef LIBCFS_VMALLOC_SIZE
154 #define LIBCFS_VMALLOC_SIZE (2 << PAGE_CACHE_SHIFT) /* 2 pages */
157 #define LIBCFS_ALLOC_PRE(size, mask) \
159 LASSERT(!in_interrupt() || \
160 ((size) <= LIBCFS_VMALLOC_SIZE && \
161 ((mask) & GFP_ATOMIC)) != 0); \
164 #define LIBCFS_ALLOC_POST(ptr, size) \
166 if (unlikely((ptr) == NULL)) { \
167 CERROR("LNET: out of memory at %s:%d (tried to alloc '" \
168 #ptr "' = %d)\n", __FILE__, __LINE__, (int)(size)); \
169 CERROR("LNET: %d total bytes allocated by lnet\n", \
170 libcfs_kmem_read()); \
172 libcfs_kmem_inc((ptr), (size)); \
173 CDEBUG(D_MALLOC, "alloc '" #ptr "': %d at %p (tot %d).\n", \
174 (int)(size), (ptr), libcfs_kmem_read()); \
179 * allocate memory with GFP flags @mask
180 * The allocated memory is zeroed-out.
182 #define LIBCFS_ALLOC_GFP(ptr, size, mask) \
184 LIBCFS_ALLOC_PRE((size), (mask)); \
185 (ptr) = (size) <= LIBCFS_VMALLOC_SIZE ? \
186 kzalloc((size), (mask)) : vzalloc(size); \
187 LIBCFS_ALLOC_POST((ptr), (size)); \
193 #define LIBCFS_ALLOC(ptr, size) \
194 LIBCFS_ALLOC_GFP(ptr, size, GFP_NOFS)
197 * non-sleeping allocator
199 #define LIBCFS_ALLOC_ATOMIC(ptr, size) \
200 LIBCFS_ALLOC_GFP(ptr, size, GFP_ATOMIC)
203 * allocate memory for specified CPU partition
204 * \a cptab != NULL, \a cpt is CPU partition id of \a cptab
205 * \a cptab == NULL, \a cpt is HW NUMA node id
206 * The allocated memory is zeroed-out.
208 #define LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, mask) \
210 LIBCFS_ALLOC_PRE((size), (mask)); \
211 (ptr) = (size) <= LIBCFS_VMALLOC_SIZE ? \
212 cfs_cpt_malloc((cptab), (cpt), (size), (mask) | __GFP_ZERO) : \
213 cfs_cpt_vzalloc((cptab), (cpt), (size)); \
214 LIBCFS_ALLOC_POST((ptr), (size)); \
217 /** default numa allocator */
218 #define LIBCFS_CPT_ALLOC(ptr, cptab, cpt, size) \
219 LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, GFP_NOFS)
221 #define LIBCFS_FREE(ptr, size) \
224 if (unlikely((ptr) == NULL)) { \
225 CERROR("LIBCFS: free NULL '" #ptr "' (%d bytes) at " \
226 "%s:%d\n", s, __FILE__, __LINE__); \
229 libcfs_kmem_dec((ptr), s); \
230 CDEBUG(D_MALLOC, "kfreed '" #ptr "': %d at %p (tot %d).\n", \
231 s, (ptr), libcfs_kmem_read()); \
232 if (unlikely(s > LIBCFS_VMALLOC_SIZE)) \
238 /******************************************************************************/
240 /* htonl hack - either this, or compile with -O2. Stupid byteorder/generic.h */
241 #if defined(__GNUC__) && (__GNUC__ >= 2) && !defined(__OPTIMIZE__)
242 #define ___htonl(x) __cpu_to_be32(x)
243 #define ___htons(x) __cpu_to_be16(x)
244 #define ___ntohl(x) __be32_to_cpu(x)
245 #define ___ntohs(x) __be16_to_cpu(x)
246 #define htonl(x) ___htonl(x)
247 #define ntohl(x) ___ntohl(x)
248 #define htons(x) ___htons(x)
249 #define ntohs(x) ___ntohs(x)
252 void libcfs_debug_dumpstack(struct task_struct *tsk);
253 void libcfs_run_upcall(char **argv);
254 void libcfs_run_lbug_upcall(struct libcfs_debug_msg_data *);
255 void libcfs_debug_dumplog(void);
256 int libcfs_debug_init(unsigned long bufsize);
257 int libcfs_debug_cleanup(void);
258 int libcfs_debug_clear_buffer(void);
259 int libcfs_debug_mark_buffer(const char *text);
261 #else /* !__KERNEL__ */
265 # define LASSERT(e) assert(e)
266 # define LASSERTF(cond, ...) \
269 CERROR(__VA_ARGS__); \
272 # define LBUG() assert(0)
273 # ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
274 # define LINVRNT(exp) LASSERT(exp)
276 # define LINVRNT(exp) ((void)sizeof!!(exp))
279 # define LASSERT(e) ((void)sizeof!!(e))
280 # define LASSERTF(cond, ...) ((void)sizeof!!(cond))
281 # define LBUG() ((void)(0))
282 # define LINVRNT(exp) ((void)sizeof!!(exp))
283 # endif /* LIBCFS_DEBUG */
284 # define KLASSERT(e) ((void)0)
285 # define printk printf
286 #define LIBCFS_ALLOC_GFP(ptr, size, mask) \
288 (ptr) = calloc(1, size); \
290 # define LIBCFS_FREE(ptr, size) do { free(ptr); } while((size) - (size))
291 # define LIBCFS_ALLOC(ptr, size) \
292 LIBCFS_ALLOC_GFP(ptr, size, 0)
293 # define LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, mask) \
294 LIBCFS_ALLOC(ptr, size)
295 # define LIBCFS_CPT_ALLOC(ptr, cptab, cpt, size) \
296 LIBCFS_ALLOC(ptr, size)
298 void libcfs_debug_dumplog(void);
299 int libcfs_debug_init(unsigned long bufsize);
300 int libcfs_debug_cleanup(void);
302 #define libcfs_debug_dumpstack(tsk) ((void)0)
305 * Generic compiler-dependent macros required for kernel
306 * build go below this comment. Actual compiler/compiler version
307 * specific implementations come from the above header files
309 #define likely(x) __builtin_expect(!!(x), 1)
310 #define unlikely(x) __builtin_expect(!!(x), 0)
314 struct cfs_cpt_table;
317 * allocate per-cpu-partition data, returned value is an array of pointers,
318 * variable can be indexed by CPU ID.
319 * cptable != NULL: size of array is number of CPU partitions
320 * cptable == NULL: size of array is number of HW cores
322 void *cfs_percpt_alloc(struct cfs_cpt_table *cptab, unsigned int size);
324 * destory per-cpu-partition variable
326 void cfs_percpt_free(void *vars);
327 int cfs_percpt_number(void *vars);
328 void *cfs_percpt_current(void *vars);
329 void *cfs_percpt_index(void *vars, int idx);
331 #define cfs_percpt_for_each(var, i, vars) \
332 for (i = 0; i < cfs_percpt_number(vars) && \
333 ((var) = (vars)[i]) != NULL; i++)
336 * allocate a variable array, returned value is an array of pointers.
337 * Caller can specify length of array by count.
339 void *cfs_array_alloc(int count, unsigned int size);
340 void cfs_array_free(void *vars);
342 #define LASSERT_ATOMIC_ENABLED (1)
344 #if LASSERT_ATOMIC_ENABLED
346 /** assert value of @a is equal to @v */
347 #define LASSERT_ATOMIC_EQ(a, v) \
349 LASSERTF(atomic_read(a) == v, \
350 "value: %d\n", atomic_read((a))); \
353 /** assert value of @a is unequal to @v */
354 #define LASSERT_ATOMIC_NE(a, v) \
356 LASSERTF(atomic_read(a) != v, \
357 "value: %d\n", atomic_read((a))); \
360 /** assert value of @a is little than @v */
361 #define LASSERT_ATOMIC_LT(a, v) \
363 LASSERTF(atomic_read(a) < v, \
364 "value: %d\n", atomic_read((a))); \
367 /** assert value of @a is little/equal to @v */
368 #define LASSERT_ATOMIC_LE(a, v) \
370 LASSERTF(atomic_read(a) <= v, \
371 "value: %d\n", atomic_read((a))); \
374 /** assert value of @a is great than @v */
375 #define LASSERT_ATOMIC_GT(a, v) \
377 LASSERTF(atomic_read(a) > v, \
378 "value: %d\n", atomic_read((a))); \
381 /** assert value of @a is great/equal to @v */
382 #define LASSERT_ATOMIC_GE(a, v) \
384 LASSERTF(atomic_read(a) >= v, \
385 "value: %d\n", atomic_read((a))); \
388 /** assert value of @a is great than @v1 and little than @v2 */
389 #define LASSERT_ATOMIC_GT_LT(a, v1, v2) \
391 int __v = atomic_read(a); \
392 LASSERTF(__v > v1 && __v < v2, "value: %d\n", __v); \
395 /** assert value of @a is great than @v1 and little/equal to @v2 */
396 #define LASSERT_ATOMIC_GT_LE(a, v1, v2) \
398 int __v = atomic_read(a); \
399 LASSERTF(__v > v1 && __v <= v2, "value: %d\n", __v); \
402 /** assert value of @a is great/equal to @v1 and little than @v2 */
403 #define LASSERT_ATOMIC_GE_LT(a, v1, v2) \
405 int __v = atomic_read(a); \
406 LASSERTF(__v >= v1 && __v < v2, "value: %d\n", __v); \
409 /** assert value of @a is great/equal to @v1 and little/equal to @v2 */
410 #define LASSERT_ATOMIC_GE_LE(a, v1, v2) \
412 int __v = atomic_read(a); \
413 LASSERTF(__v >= v1 && __v <= v2, "value: %d\n", __v); \
416 #else /* !LASSERT_ATOMIC_ENABLED */
418 #define LASSERT_ATOMIC_EQ(a, v) do {} while (0)
419 #define LASSERT_ATOMIC_NE(a, v) do {} while (0)
420 #define LASSERT_ATOMIC_LT(a, v) do {} while (0)
421 #define LASSERT_ATOMIC_LE(a, v) do {} while (0)
422 #define LASSERT_ATOMIC_GT(a, v) do {} while (0)
423 #define LASSERT_ATOMIC_GE(a, v) do {} while (0)
424 #define LASSERT_ATOMIC_GT_LT(a, v1, v2) do {} while (0)
425 #define LASSERT_ATOMIC_GT_LE(a, v1, v2) do {} while (0)
426 #define LASSERT_ATOMIC_GE_LT(a, v1, v2) do {} while (0)
427 #define LASSERT_ATOMIC_GE_LE(a, v1, v2) do {} while (0)
429 #endif /* LASSERT_ATOMIC_ENABLED */
431 #define LASSERT_ATOMIC_ZERO(a) LASSERT_ATOMIC_EQ(a, 0)
432 #define LASSERT_ATOMIC_POS(a) LASSERT_ATOMIC_GT(a, 0)
434 #define CFS_ALLOC_PTR(ptr) LIBCFS_ALLOC(ptr, sizeof (*(ptr)));
435 #define CFS_FREE_PTR(ptr) LIBCFS_FREE(ptr, sizeof (*(ptr)));
438 * percpu partition lock
440 * There are some use-cases like this in Lustre:
441 * . each CPU partition has it's own private data which is frequently changed,
442 * and mostly by the local CPU partition.
443 * . all CPU partitions share some global data, these data are rarely changed.
445 * LNet is typical example.
446 * CPU partition lock is designed for this kind of use-cases:
447 * . each CPU partition has it's own private lock
448 * . change on private data just needs to take the private lock
449 * . read on shared data just needs to take _any_ of private locks
450 * . change on shared data needs to take _all_ private locks,
451 * which is slow and should be really rare.
455 CFS_PERCPT_LOCK_EX = -1, /* negative */
460 struct cfs_percpt_lock {
461 /* cpu-partition-table for this lock */
462 struct cfs_cpt_table *pcl_cptab;
463 /* exclusively locked */
464 unsigned int pcl_locked;
465 /* private lock table */
466 spinlock_t **pcl_locks;
469 /* return number of private locks */
470 #define cfs_percpt_lock_num(pcl) cfs_cpt_number(pcl->pcl_cptab)
473 * create a cpu-partition lock based on CPU partition table \a cptab,
474 * each private lock has extra \a psize bytes padding data
476 struct cfs_percpt_lock *cfs_percpt_lock_alloc(struct cfs_cpt_table *cptab);
477 /* destroy a cpu-partition lock */
478 void cfs_percpt_lock_free(struct cfs_percpt_lock *pcl);
480 /* lock private lock \a index of \a pcl */
481 void cfs_percpt_lock(struct cfs_percpt_lock *pcl, int index);
482 /* unlock private lock \a index of \a pcl */
483 void cfs_percpt_unlock(struct cfs_percpt_lock *pcl, int index);
484 /* create percpt (atomic) refcount based on @cptab */
485 atomic_t **cfs_percpt_atomic_alloc(struct cfs_cpt_table *cptab, int val);
486 /* destroy percpt refcount */
487 void cfs_percpt_atomic_free(atomic_t **refs);
488 /* return sum of all percpu refs */
489 int cfs_percpt_atomic_summary(atomic_t **refs);
490 #endif /* __KERNEL__ */
492 /** Compile-time assertion.
494 * Check an invariant described by a constant expression at compile time by
495 * forcing a compiler error if it does not hold. \a cond must be a constant
496 * expression as defined by the ISO C Standard:
498 * 6.8.4.2 The switch statement
500 * [#3] The expression of each case label shall be an integer
501 * constant expression and no two of the case constant
502 * expressions in the same switch statement shall have the same
503 * value after conversion...
506 #define CLASSERT(cond) do {switch (1) {case (cond): case 0: break; } } while (0)
509 #define ergo(a, b) (!(a) || (b))
510 /* logical equivalence */
511 #define equi(a, b) (!!(a) == !!(b))
513 #ifndef CFS_CURRENT_TIME
514 # define CFS_CURRENT_TIME time(0)
517 struct libcfs_device_userstate
519 int ldu_memhog_pages;
520 struct page *ldu_memhog_root_page;
523 /* what used to be in portals_lib.h */
525 # define MIN(a,b) (((a)<(b)) ? (a): (b))
528 # define MAX(a,b) (((a)>(b)) ? (a): (b))
531 #define MKSTR(ptr) ((ptr))? (ptr) : ""
533 static inline size_t cfs_size_round4(size_t val)
535 return (val + 3) & (~0x3);
538 #ifndef HAVE_CFS_SIZE_ROUND
539 static inline size_t cfs_size_round(size_t val)
541 return (val + 7) & (~0x7);
543 #define HAVE_CFS_SIZE_ROUND
546 static inline size_t cfs_size_round16(size_t val)
548 return (val + 0xf) & (~0xf);
551 static inline size_t cfs_size_round32(size_t val)
553 return (val + 0x1f) & (~0x1f);
556 static inline size_t cfs_size_round0(size_t val)
560 return (val + 1 + 7) & (~0x7);
563 static inline size_t cfs_round_strlen(char *fset)
565 return cfs_size_round(strlen(fset) + 1);
568 /* roundup \a val to power2 */
569 static inline size_t cfs_power2_roundup(size_t val)
571 if (val != LOWEST_BIT_SET(val)) { /* not a power of 2 already */
573 val &= ~LOWEST_BIT_SET(val);
574 } while (val != LOWEST_BIT_SET(val));
575 /* ...and round up */
581 #define LOGL(var,len,ptr) \
584 memcpy((char *)ptr, (const char *)var, len); \
585 ptr += cfs_size_round(len); \
588 #define LOGU(var,len,ptr) \
591 memcpy((char *)var, (const char *)ptr, len); \
592 ptr += cfs_size_round(len); \
595 extern struct cfs_psdev_ops libcfs_psdev_ops;
596 extern struct miscdevice libcfs_dev;
597 extern struct cfs_wi_sched *cfs_sched_rehash;
599 extern int insert_proc(void);
600 extern void remove_proc(void);